International Journal of Public Opinion Research Vol.  No. 
–/ $.
ISSUE ATTRIBUTES AND
AGENDA-SETTING BY MEDIA, THE
PUBLIC, AND POLICYMAKERS IN
CANADA
Stuart N. Soroka
ABSTRACT
Agenda-setting hypotheses inform research on both media influence and policy making.
The study draws from these two literatures, building a more accurate and comprehensive
model of the expanded agenda-setting process. Evidence is derived from a longitudinal
dataset, including a content analysis of Canadian newspapers, results from public
opinion polls, and measures of attention to issues in Question Period, committees,
Throne Speeches, and legislative initiatives from  to . A model is estimated
that accommodates dynamic, multi-directional effects. Findings are presented for three
issues—inflation, environment, and debt/deficit—with an eye on examining different
agenda-setting dynamics, and the degree to which these dynamics are linked to issue
attributes. The results () demonstrate the value of an agenda-setting framework and
a means of modelling media effects and the policy making process, and () indicate the
importance of taking issue attributes into account in predicting or accounting for
agenda-setting effects.
Relationships between mass media, the public, and policymakers are at the
centre of both political communications and everyday politics. These interactions
reach into a wide range of research interests, including the media’s role in the
formation of public opinion and public policy, and the degree to which public
policy follows or leads public opinion. Abstractly, we might reflect on the wider
implications these relationships have for democratic theory or institutional
development. Concretely, we might consider what they tell us about day-to-day
This article was first submitted to IJPOR April , . The final version was received July , .
The author wishes to thank Richard Johnston, Byron Shafer, and the Journal’s anonymous referees for their
comments on previous versions of this paper, Donald Blake, Bryan Jones, Richard Jenkins, and Patrick
Fournier for their helpful comments on related work, and William Veloce and Lewis Soroka for their help
with economic and statistical matters. The Social Sciences and Humanities Research Council of Canada
provided support for this research in the form of a doctoral fellowship. Public opinion data analysed within
were made available by the University of British Columbia Numeric Data Services, the Centre for the Study
of Democracy at Queen’s University, and the Carleton University Data Centre. None of these institutions
or individuals bears any responsibility for errors or omissions in the final product.
 World Association for Public Opinion Research 
   -  

interactions between newspapers, television, citizens, politicians, and bureaucrats.
That these topics are typically examined only in part is no surprise, considering
the scope of the territory covered. An agenda-setting framework offers the
possibility of looking concurrently at a wide range of political relationships,
however, and of empirically mapping political communications at the societal
level. Admittedly, agenda-setting research has rarely capitalized on this potential.
Work on public agenda-setting has seldom drawn from policy agenda-setting
research, and vice versa. Nevertheless, past work suggests that agenda-setting
models are capable of accommodating both media-public dynamics and the
relationships between these actors and the policy process.
The paper that follows represents one attempt at doing exactly this. It is
based on data for three issues from Canada, but the model is easily adapted to
any issue in any democracy. Policy measures will likely change with different
governing institutions, of course, as will media-public-policy relationships.
But the general tenets upon which this research is based—the value of an
agenda-setting framework, a statistical model of the expanded agenda-setting
process, and an acknowledged relationship between agenda-setting dynamics
and issue attributes—are described here with the expectation that they are
valuable beyond the Canadian examples used below.
We begin with a brief review of public and policy agenda-setting research.
This body of research is considerable, as are its contributions to political science
and communications. Nevertheless, the literature has remained splintered and
is often beleaguered by methodological difficulties. Some of these difficulties
are described below, and a new model and measures are presented with the aim
both of uniting the vast bodies of past public and policy agenda-setting work,
and of better accommodating dynamic, multi-directional relationships. This
model is then estimated for three issues in Canada—inflation, environment,
and debt/deficit. Results are discussed as they pertain to the Canadian experience
in particular and, more generally, to the agenda-setting framework and the
connection between issue attributes and issue dynamics.
AGENDA-SETTING: PROBLEMS AND POTENTIAL
Agenda-setting research focuses not on issue opinions, per se, but on issue
salience. Cohen was the first to state what has become the central public
agenda-setting hypothesis: the press ‘may not be successful much of the time
in telling people what to think, but it is stunningly successful in telling its
readers what to think about’ (Cohen , p. ). Following from Cohen’s
discussion, public agenda-setting work demonstrates that increased issue salience
for the media leads to increased issue salience for the public—in agenda-setting
terms, that the media agenda has an impact on the public agenda, where an

     
agenda is a ‘ranking of the relative importance of various public issues’ (Dearing
, p. ). McCombs and Shaw’s () study stands as the first empirical
public agenda-setting study; well over  have followed in their footsteps.1
Cobb and Elder (), meanwhile, have used a similar framework to examine
public policy formation. Subsequent policy agenda-setting studies examine relationships between media and policy, as well as public and policy agendas.2
Studies by Kingdon () and Baumgartner and Jones () represent the
current state of this line of research; in their consideration of public opinion, in
fact, Baumgartner and Jones’ work stands as one of the few recent policy-oriented
studies to draw from both the public and policy agenda-setting traditions.
The combination of empirical work on public opinion and on public policy
is long overdue; recent efforts include only half-hearted attempts to build
models using both public opinion and policy measures. Nevertheless, it is clear
that effects of/on the public and of/on policymakers are intimately connected,
and the most significant advantage of an agenda-setting framework is its
ability—through the use of a common vernacular and directly comparable
measures—to combine mass media analysis, public opinion research, and studies
of the policymaking process. In this way, agenda-setting work is uniquely
qualified to offer empirical accounts of political communications at the societal
level. A central goal of the current work is to demonstrate this fact, by making
an explicit effort to combine public opinion- and policy-oriented agenda-setting
analysis, and by building a model that empirically links media, public, and
policy agendas.
Combining disparate literatures on agenda-setting is the first goal of this
project; recognizing and accommodating for the fact that different issues will
have different agenda-setting dynamics is the second. Different issues have led
to markedly different agenda-setting results, after all, and recent work suggests
that the direction of media-public, media-policy, and public-policy relationships
vary both across issues and over time (Brosius and Kepplinger , Gonzenbach
, Soroka a). Different hypotheses have been offered to account for this
varied evidence. Some is due simply to methodological differences—measures of
agendas and means of analysis have varied widely. Issue attributes also play a
role, however. Zucker’s () ‘obtrusiveness’ hypothesis is perhaps the bestknown issue attribute theory: he suggests that the more obtrusive an issue
is—the more likely individuals experience it directly—the less potential there
is for media effects on public opinion. Other authors have suggested additional
hypotheses: () ‘concrete’ issues should be more open to media effects than
1
For thorough reviews of the agenda-setting literature, see Dearing and Rogers (), McCombs and
Shaw (), and Rogers et al. ().
2
There is no equivalent review of the policy agenda-setting literature, although some of it is covered in
the public agenda-setting reviews. For a representative sample of work on the USA—other than those
mentioned above—see Mayer (), Page and Shapiro (, ), Pritchard (), and Wanta et al.
(); for work on Canada, see Howlett (, ) and Soroka (b, ).
   -  

‘abstract’ issues (Yagade and Dozier ); () the public has a limited attention
span, so issues that are salient for a long period will eventually offer less
opportunity for media impact (Downs , Zucker ); () issues that
involve dramatic events or conflict should have an increased potential for media
attention and effects on public opinion (MacKuen and Coombs , Wanta
and Hu ).
In spite of their potential for accounting for (and predicting) variation
in agenda-setting effects, hypotheses about the role of issue attributes in
agenda-setting have received only intermittent attention in the public agendasetting literature. Moreover, they have tended to deal more with the media-public
link than with links between these and policy agendas. In fact, no effort has
been made to test hypotheses regarding the role of issue attributes in the larger
agenda-setting process. Accordingly, three issues—inflation, environment, and
debt/deficit—are examined below with an eye towards both highlighting differences in issue attributes, and testing the potential for issue attribute hypotheses
in accounting for different media-public-policy agenda-setting dynamics.3
THE ISSUES
I
Inflation in Canada from  to  falls into two periods. ‘From  to
’, Theissen () writes, monetary policy in Canada was carried out
with price stability as the longer-term goal and inflation containment as the
shorter-term goal, but without intermediate targets or a specified path to the
longer-term objective’ (Theissen , p. ). This method proved unsatisfactory, leading to a peak in inflation in –. New inflation targets
were then introduced as part of the  budget, and inflation began to decline
dramatically. In fact, actual inflation declined much more quickly than expected
inflation during this period (Johnson ), reaching a low point in  and
then balancing out for the rest of the period.
We expect inflation to be the quintessential ‘obtrusive’ issue. This is an issue
that most individuals experience regularly—the public does not need the media
to tell them when inflation is bad. Rather, public concern about inflation should
simply follow the rate of inflation. Indeed, Figure , along with the bulk of
past work on public agenda-setting for inflation shows that this is an accurate
description of trends in public opinion (Behr and Iyengar , Demers et al.
3
The issues surveyed here represent a subset of the eight issues originally investigated by the author:
AIDS, crime, debt/deficit, environment, inflation, national unity, taxes, and unemployment (see Soroka
a, ). These three issues have been selected since they illustrate variation in both issue attributes
and agenda-setting results; it is worth noting, however, that the theory and methodology used here was
successful at modeling agenda-setting dynamics for seven of the eight issues originally surveyed. (Due to
little variation in public opinion on AIDS, this estimation was less successful.)

     
, MacKuen and Coombs , Winter et al. ). The forthcoming
estimation will provide an additional test of this hypothesis.
E
Environmental issues went through a period of increased salience in the late
s/early s. This period is marked by the crash of the Exxon Valdez oil
tanker in March , the twentieth anniversary of Earth Day in April ,
and the much-touted June  United Nations Conference on Environment
and Development in Rio de Janeiro. Environmental policy-making was also at
a high during this period, with the Canadian Environmental Protection Act
() and the formulation and introduction of the Green Plan (). The
Plan detailed federal environmental initiatives and objectives, and served as a
guide for Federal behaviour until .
Environment should be more open to media influence than inflation. Most
individuals will not experience environmental problems everyday—the issue is
‘unobtrustive’, so there exists the possibility that the media will seize on an
environmental issue and increase public concern. Additionally, the existence of
dramatic events should increase the potential for media effects. Environmental
issues have received considerable attention in agenda-setting literature, partly
because they play the role of guinea pig in Downs’ () seminal ‘issue
attention cycle’ article. Again, both the time series illustrated in Figure  and
past research suggest a similar result: the media agenda should lead the public
and policy agendas, and there should be virtually no relationship between any
of these agendas and real-world environmental indicators (Ader , Atwater
et al. , Brosius and Kepplinger , Hansen , Hester and Gonzenbach
, Iyengar and Kinder , MacKuen and Coombs , Parlour and
Schatzow , Smith ).
D  D
After declining gradually since World War II, Canadian federal government
debt (as a percentage of GDP) began to climb again in the mid-s. This
was largely a product of the expansionary fiscal policy used to fight effects of
the first oil shock in the early s and the recession in the early s. The
federal government was extremely lax in controlling its debt throughout the
s (Kneebone ). Dodge suggests that political and public attention to
the debt was restricted during this time by the focus on free trade and tax
reform: ‘Governments and the public can really only focus on one or two
economic issues at a time’ (Dodge , p. ). As a result, public attention
and political will did not grow until the debt problem had worsened considerably.
Politicians and the public recognized it as a major problem only passingly in
   -  

April of , the month of the first budget discussion in which deficit and
deficit slashing were identified as pressing concerns, and then not for an
extended period until  (with the budget of that year, the following election,
and actions by the new Liberal government to eliminate deficits). The height
of concern over federal debt and deficits occurred after the period surveyed
here, likely around the time of the  election.
The debt/deficit is both unobtrusive and abstract, and is unlikely to become
salient unless policymakers emphasize it. The public does not feel the presence
of the federal debt/deficit directly, after all, and the media are likely to remain
either oblivious to or uninterested in the problem until policymakers highlight
it. Past agenda-setting research has demonstrated moderate effects of the media
on public attention to public debt and deficits (Brosius and Kepplinger ,
Jasperson et al. , Weaver ). These studies have not considered the role
of policymakers, however, and in the Canadian case—where both periods of
heightened debt/deficit salience seem to have been brought on by government
actors identifying the problem and staking a political claim—the policy agenda
is likely an important part of the story. Accordingly, the increased salience of
debt/deficit in Canada should provide an example of policy agendas leading
media and public agendas.
In short, past issue attribute hypotheses suggest that inflation should be
real-world-driven, environment should be media-driven, and debt/deficit should
be policy-driven. The modeling effort that follows represents one attempt to
test these specific hypotheses. More generally, our work is guided by the
following expectations: () different issues display different agenda-setting
dynamics, and () this variance in dynamics is linked to issue attributes.
RESEARCH DESIGN
T M
Past agenda-setting work has shown that relationships between the media,
public, and policymakers can vary in direction, or be bi-directional. That the
statistical procedures used have not been able to accommodate or estimate these
dynamic, bi-directional links, then, stands as a major flaw with much past work.
Bartels’ () recent examination of the press-policy relationship stands as an
exception—he uses vector autoregression (VAR) to model the bi-directional
relationship between these two agendas. The work below follows suit by
estimating a system of equations using seemingly unrelated regressions (SUR).
We begin with an illustration of the expanded model upon which our equations
are based. Figure  illustrates the hypothesized relationships between the media,
public, and policymakers. The three primary agendas are illustrated, and causal
arrows run in both directions between most of them. There is no direct link

     
F  An expanded model of the agenda-setting process
from the policy to the public agenda, based on the assumption that policymakers
can affect the public through the media or real-world indicators, but not directly.
Real-world factors are located at the centre, affected by policymakers and
affecting each of the three agendas. A number of sub-agendas are illustrated
within each of the primary agendas. This points to the fact that the primary
agendas might be represented in different ways, but also to the possibility of
both intra- and inter-agenda analysis. We might explore whether the New York
Times affects the media agenda, for example, or the relationship between the
Cabinet and House agendas. The model, then, accomplishes two tasks: () it
provides a framework with which to compare, contrast, and combine a wide
variety of agenda-setting analyses, and () it makes few assumptions about
directions of causality, emphasizing the possibility of multi-directional agendasetting and suggesting a structure for empirical investigation.
Following from Figure , an expanded agenda-setting process is estimated
below using a system of three equations—one for each of the primary agendas.
The basic model is as follows:
MDt=
MDt−k+PBt−k+QPt−k+Cmt,t++PMbt,t++Gbt,t++THSpt,t++Ect−,t,t+
+Bdt−,t,t++RWt−m
()
PBt=MDt−k+PBt−k+Ect−,t,t++RWt−m
()
QPt=
MDt−k+PBt−k+QPt−k+Cmt,t++PMbt,t++Gbt,t++THSpt,t++Bdt−,t,t+
+RWt−m+sit,
()
where MD is the media agenda, PB is the public agenda, QP is Question Period
content, Cm is Parliamentary committee reports, PMb is Private Members’
Bills, Gb is Government Bills, THSp is Throne Speech content, Ec, Bd, and
   -  

sit are dummy variables representing elections, budgets, and whether the House
was sitting in a given month, and RW is real-world indicators.
T M
MD, PB, and QP are the three endogenous variables. The Canadian media
agenda is measured using title searches for seven English newspapers (the
Halifax Chronicle Herald, Montreal Gazette, Globe and Mail, Toronto Star,
Winnipeg Free Press, Calgary Herald, and Vancouver Sun) in Canadian Business
and Current Affairs, and one French newspaper (La Presse) in Eureka. The end
result is a monthly time-series of the number of articles in Canadian newspapers
with a particular keyword (e.g., ‘inflation’) in the title.4
The public opinion time series is based on responses to an open-ended survey
question similar to, ‘What do you think is the most important problem facing
Canada today?’, every time it was asked by six Canadian polling firms (Angus
Reid, CBC/Globe and Mail, Decima, Environics, Gallup, and Pollara). While
question wording varies somewhat, Smith (, ) suggests that most
responses to ‘the most important problem’ (MIP) question are generally not
affected by wording changes; preliminary tests led Soroka () to the same
conclusion. Results from the six firms are roughly comparable, then, and the
responses are combined to create a monthly measure of the proportion of
respondents giving a particular response (e.g. ‘inflation’).5
While the specification of media and public agendas is relatively straightforward, selecting a single measure of the policy agenda is more difficult. There
is a considerable body of literature dealing with the policy agenda, certainly,
but its measures vary widely. Moreover, the vast majority of work on policy
agendas has been performed in the USA, where committee meetings, bill
introductions, and presidential papers are easily measured and stand as reasonably
accurate indicators of the policy agenda. The situation is very different in
Canada, where a cabinet-centred parliamentary system with extraordinarily
4
The newspapers used provide a good sample of newspapers in different provinces. La Presse is the only
French-language newspaper for which data are available from . All time series are the product of
title searches using the following keywords: Inflation: inflation. Environment: environment; environmental;
environmentalist; environmentalism; conservation; conservationalist; ozone; endangered and species; endangered and animal; endangered and plant; endangered and tree; endangered and fish; global warming;
clearcut; clearcutting; acid rain; pollution; pollute; pollutant; polluter. Debt/Deficit: debt and national; debt
and federal; debt and government; debt and public; deficit and national; deficit and federal. (Plurals are not
listed, but were included in searches where appropriate. French equivalents, and feminine forms of words,
were used for La Presse in Eureka.)
5
In cases when more than one poll exists for a given month, results are combined and the proportion of
all responses is used. A limited number of cases allowed for two responses. In these cases, again, the
proportion of all responses is used (as in Behr and Iyengar ). Data are interpolated, using linear
interpolation, for months in which no poll exists; this did not have to be done for very many months, and
never more than two consecutively. Variations in question wording and the specific dates of polls are not
listed here, but are available in Soroka (a, ), or upon request from the author.

     
strong party discipline ensures that committees play a small role, the potential
for bill introductions is severely restricted, and the majority of important policy
discussion takes place unrecorded behind closed doors.
Taking these factors into account, it is likely that debate in the legislature is
the best single measure of the policy agenda in Canada. Question Period (QP)—a
forty-minute period during which opposition parties pose questions to Cabinet
members—seems especially likely to offer an indication of the issues that are
important to policymakers on a daily basis. Moreover, QP appears to be one
governmental forum where day-to-day changes may be evident. Pritchard and
Berkowitz (, p. ) suggest that symbolic agendas, relating primarily to
attention by policymakers to issues, are more easily changed than resource
agendas, which are more closely linked to the allocation of capital (human or
financial). We should expect, then, that while spending priorities or legislative
initiatives will be slow to show agenda change, QP will be more likely to show
short-term changes in issue salience.
The susceptibility of QP to short-term change is especially useful for a study
using ten years of monthly data. Accordingly, QP content is used here as our
primary measure of the policy agenda: QP is the number of column centimetres
in Hansard, monthly, during Question Period, dealing with one of our three
issues. The number of column centimetres is divided by the number of days
the House sat each month to control for differences in this regard, and a dummy
variable for months in which the House did not sit at all (sit) is included in
equation  to soak up the effects of months in which there was no potential for
the QP measure to be anything other than zero.
We must be careful not to treat QP content as an indicator of the entire
policy agenda, of course—QP content is an accurate indication of what MPs
are discussing in Question Period, but it is far less accurate at indicating
legislative initiatives, committee discussions, or changes in spending priorities.
That said, in an effort to better account for the possibility that media content
and/or public opinion are affected by these other indicators of policy agendas,
several additional measures are included as exogenous variables: () the number
of related committee reports presented in the House each month (Cm), () the
number of bill discussions (Private Members’ and Government Bills; st, nd,
rd, and other readings) in the House each month, and () the proportion of
the Throne Speech given to an issue in months when there was a Throne
Speech (THSp).6 The fact that these are more resource-oriented (and will
therefore tend to be more incremental in change than Question Period),
justifies—or at least necessitates—their exogeneity in our model of relatively
short-term effects. Nevertheless, including these measures provides a better
6
The Throne Speech is the Canadian equivalent of the State of the Union Address in the USA, or the
Queen’s Speech in the UK. It is a statement of the Government’s priorities, read by the Governor General,
at the beginning of a new session of Parliament.
   -  

representation of the policy agenda and should improve our ability to identify
effects of policy agendas on media content and public opinion.
The model also includes dummy variables for elections and budgets (Ec, Bd).
These events may have an effect on attention to issues, and so are included
here both to test this possibility and to control for these effects in the larger
estimation. Since the House does not usually sit for several months previous
to or following an election, the election variable is included only in the media
and public equations. Budgets, on the other hand, should not affect the public
directly—most people do not have direct experience with budgets, so effects of
the budget, and other policy indicators, on public opinion should occur via the
media. Accordingly, this variable is included in only the media and policy
equations.
Finally, the models include real-world indicators (RW). The inflation model
includes monthly changes in the consumer price index (CPI); the debt/deficit
model includes yearly changes in the debt as a proportion of GDP, and yearly
changes in the deficit as a proportion of Federal Government spending; the
environment model includes yearly changes in the number of hectares harvested
(forests), ozone depleting substances, carbon dioxide emissions, and number of
species at risk.7 The multiple environmental measures reflect the difficulty in
selecting a single measure of the state of the environment. Nevertheless,
including real-world indicators in the model is essential to the overall estimation.
On the one hand, including real-world indicators allows us to judge the degree
to which various agendas are connected to the real world. Perhaps more
fundamental, however, is the fact that evidence of relationships between media,
public, or policy agendas is more convincing if we can show that these
relationships function independently of real-world circumstances.8 Differenced
series are used based on a belief that reactions by the public, media, and
policymakers will most often be to relative changes in the various real world
indicators rather than their actual values.
The number of lags used for endogenous variables (k) and real world indicators
(m) is determined using the usual strategy for VAR/SUR estimations: () the
model is estimated using a maximum number of lags (based either on data
constraints or theoretical premise), () lags are dropped one-by-one if they are
not statistically significant, and () provided the model’s residuals show no
autocorrelation, the model is accepted. The number of lags for exogenous
7
Details on real-world indicators are as follows: Consumer price indexes (monthly,  classification),
Debt (annually, as a proportion of GDP), and Deficit (annually, as a proportion of Federal Government
Revenue) are from Statistics Canada; timber harvest levels (Annual area harvested, s of hectares), new
supplies of ozone-depleting substances, CFCs and other ODS, Carbon dioxide emissions from fossil fuel use
(megatonnes), and cumulative change in risked species (of all species, subspecies, and populations evaluated
by COSEWIC) are reported in Canada’s Environmental Indicators Series, Environment Canada.
8
For further discussion about the importance of real world indicators in agenda-setting estimations, see
Behr and Iyengar (), Rogers and Dearing (), and Zucker ().

     
variables was based on a series of preliminary tests. Any correlation between
the exogenous policy variables and the others exists either concurrently or
within a one-month period. Accordingly, they are included at lags t and t+
only. Any election and budget effects also happen within a one-month period,
but the impact of these phenomena sometimes precedes the actual event.9
Election and budget variables, then, are included at lags t−, t, and t+.
Media, policy, and real-world data were collected from  to . Not
all issues were coded by all polling firms from  onwards, however, so the
estimations are restricted to the time-period for which public opinion data were
available. The inflation estimation begins in May ; environment in June
; debt/deficit in May .
T E
Equations  to  are estimated as a system using Zellner’s () seemingly
unrelated regressions (SUR) procedures.10 The only problem with solving the
equations in this way is that multiple lags and an estimation that allows for
both direct and indirect effects combine to make interpreting the results
considerably more difficult than in a standard regression. The multicollinearity
that can result from the use of many lagged variables, for example, produces
the possibility that, while the overall effects are accurate, the individual coefficients are not. We can no longer examine individual coefficients; we can,
however, make more convincing claims about causal relationships between
variables (Freeman et al. ).
Accordingly, our interpretation of the results is based on Granger exogeneity
tests and impulse response functions. These are typical means of interpreting
VAR/SUR models; the methodology involved, and their use in political science,
has been well described elsewhere. Granger tests, on the one hand, are used to
test the null hypothesis that the sum of all lags for a given variable in a single
equation is equal to zero.11 These tests provide a snapshot of causal effects;
they do not indicate the sign of the effects, however, and give only a thin
9
The effect of elections on the public, for instance, can happen during the campaign period in the month
preceding the actual vote. Similarly, anticipation of the April budget can have an effect on Question Period
discussion in March.
10
Statistically speaking, SUR techniques use Aitken’s generalized least squares (GLS) to solve a system
of equations accounting for linked error terms, creating more efficient estimates in certain situations than
either ordinary least squares (OLS) on individual equations or a vector autoregression (VAR) OLS estimation.
SUR is used here rather than VAR because the right-hand side variables are not the same across equations
 to . For a discussion of the advantages of SUR over VAR when this is the case, see Greene () or
Dwivedi and Srivastava ().
11
This is an extension of the simple Granger causality test (Granger , Freeman ): () a model
such as equation  is estimated with and without variable PBt, and () a likelihood ratio test is used to test
the null hypothesis that the history of PBt contributes nothing to the prediction of MDt. A statistically
significant test is taken as an indication of causality, while an insignificant test is taken as an indication that
PBt is exogenous to the process.
   -  

indication of the dynamic relationship between two variables. Impulse response
functions, then, are used to provide more detailed information about the
relationships between endogenous variables. The mechanics are complicated
and will be left to statistics texts (e.g. Enders ); for our purposes, a brief
explanation will suffice. All autoregressive (AR) processes have a moving average
(MA) representation, where the current value of an endogenous variable is
expressed in terms of present and past shocks to each endogenous variable. For
a VAR/SUR estimation, the resulting sets of MA coefficients are called impulse
response functions, and plotting them over time is a way to visually represent
the behaviour of the endogenous series in response to shocks in one of these
series. If results are standardized, the effects of or on each variable are directly
comparable, and provide a useful tool with which to examine the magnitude,
direction, and duration of causal effects.12
In sum, our model is aimed at measuring multi-directional causal relationships
between media, public, and policy agendas. Efforts have been made to include
all the relevant variables and causal links. Statistically speaking, this should
increase the accuracy of our estimations, and lead to more reliable indications
of causality. Practically speaking, it should lead to a more complete picture of
political communications in Canada, and provide a model that is easily adapted
to a wide variety of countries and issues.
RESULTS
Results for all issues are displayed in Table  and Figure . Table  (A to C)
presents the results of the Granger exogeneity tests. Each dependent variable
is listed (column ) along with all its independents (column ). The chi-square
test (column ) tests the null hypothesis that all the coefficients for that
independent variable are not different than zero. A significant chi-square test
(column ), therefore, indicates that lags of the independent variable contribute
significant information to the prediction of the dependent, above and beyond
the effects of all the other variables in the system, and—since this is an SUR
estimation—taking into account the possibility that the individual equations are
related.
Figure  (A to I) illustrates the impulse response functions. In each case,
three graphs are displayed for each issue. Those in the first column illustrate
the effects on each agenda of a one standard deviation impulse in the media
agenda; graphs in the next two columns do the same for the public and policy
12
Impulse response functions are estimated using RATS (Regression Analysis for Time Series), using a
relatively simple Bernanke–Sims decomposition (Bernanke , Sims ), where each series is affected
contemporaneously by its own innovations, but effects of the other two endogenous series do not appear
until the next month.

     
T  Granger exogeneity test results
Chi a
Sig.
A Inflation
Media Media
Public
Policy
Throne Speech
Elections
Budget
Real World
.
.
.
.
.
.
.
(.) Public
(.)
(.)
(.)
(.)
(.) Policy
(.)
Media
Public
Election
Real World
B Environment
Media Media
Public
Policy
Committees
Bills
Throne Speech
Election
Budget
Real World
.
.
.
.
.
.
.
.
.
(.) Public
(.)
(.)
(.)
(.)
(.) Policy
(.)
(.)
(.)
C Debt and deficit
Media Media
Public
Policy
Committees
Bills
Throne Speech
Elections
Budget
Real World
.
.
.
.
.
.
.
.
.
(.) Public
(.)
(.)
(.)
(.)
(.)
(.) Policy
(.)
(.)
DeIndependent
pendent variable
variable
Chi square tests in bold are significant at p < .
a
df=. b df=.
Chi a
Sig.
.
.
.
.
(.)
(.)
(.)
(.)
Media
Public
Policy
Throne Speech
Budget
Real World
.
.
.
.
.
.
(.)
(.)
(.)
(.)
(.)
(.)
Media
Public
Election
Real World
.
.
.
.
(.)
(.)
(.)
(.)
Media
Public
Policy
Committees
Bills
Throne Speech
Budget
Real World
.
.
.
.
.
.
.
.
(.)
(.)
(.)
(.)
(.)
(.)
(.)
(.)
Media
Public
Election
Real World
Economy
.
.
.
.
.
(.)
(.)
(.)
(.)
(.)
Media
Public
Policy
Economy
Committees
Bills
Throne Speech
Budget
Real World
Economy
.
.
.
.
.
.
.
.
.
.
(.)
(.)
(.)
(.)
(.)
(.)
(.)
(.)
(.)
(.)
DeIndependent
pendent variable
variable
   -  

F  Issue time series
agendas, respectively. These graphs are discussed below, along with the Granger
results, issue by issue.
Granger test results for inflation (Table A) offer an obvious and strong
illustration of prominent issue dynamics. Each series is autocorrelated—past
values of each series are significant predictors of current values (the media
coefficients are significant in the media model, for example). More importantly,
significance tests of real-world lags indicate that the public and policy agendas
are led by changes in the CPI. Some other effects are also evident—budgets
spark attention to inflation in the media, for example, while Throne Speech
content creates a similar rise in attention during Question Period. There is also
a significant effect of the public on QP content, and the policy on the media
agenda. Impulse response functions (Figure ) indicate that the public impact

     
F  Impulse response functions
on QP content is negative, however, and so has no substantive explanation.
The dominance of the real-world indicator over other inter-agenda effects is
also illustrated in the impulse response functions (Figure A–C). Public opinion
appears to be virtually independent of the media agenda—an impact on the
media series leads to only a very small rise in the public series in periods –
(Figure A). There is a media-policy interaction, with a Question Period impact
on media at period , and a slightly more sizeable media impact on the public
agenda at periods – (Figure C, A). Finally, it is worth noting that public
opinion has a moderate positive effect on media content at periods – (Figure
B). This may suggest that the public reacts early to increases in inflation, and
   -  

that media response is dually to inflation and to public reaction to it. The
magnitude of all these effects is relatively small, however; the absence of
significant effects in the impulse response functions illustrates the fact that
concern about inflation is driven for the most part by inflation itself.
In contrast, the Granger tests for environment (Table B) present evidence
of sensational issue, media-driven effects. Again, each endogenous series is
autocorrelated. Unlike results for inflation, however, environment results show
a significant impact of the media on both the public and policy agendas. This
is also reflected in the impulse response functions—a rise in media attention
leads to increases in both the public and policy series (Figure D). By way of
comparison, the effects of the media on the public agenda are roughly twice as
large as those for inflation, and considerably more durable.
That said, media effects are only part of the story. The public agenda affects
the policy agenda at period , and has a delayed but considerable impact on
the media agenda as well. The policy agenda also has an effect on media
attention at period  (Figure E, F). These effects are also evident in Granger
tests, where all three endogenous agendas are significant predictors of all
other agendas. The media plays a significant role in increased attention to
environmental issues, therefore, but results indicate that the media, public, and
policymakers play mutually reinforcing roles.
While uni-directional effects of the media on other agendas are not identified
in this model, however, additional considerations do suggest the comparative
significance of the media as a leader over the public and policy agendas. Impulse
response functions suggest that media effects on the public begin at period 
and are lasting, for instance. Public effects on the media are initially small and
negative, on the other hand, and do not become positive until period four
(Figure D, E). These dynamics suggest the possibility that while the relationship
between the media and public becomes a reciprocal one, the initial relationship
is one in which the media leads.
Impulse response functions also indicate that media effects on policymakers
are slightly greater and more prolonged than are effects in the other direction.
The predominant direction of the media-policy relationship can be determined
in a more convincing manner using weekly series, however. While these
estimations use monthly data, both the media and policy time series are available
on a weekly basis, and similar analyses of these data may provide a more
accurate look at inter-agenda dynamics. Granger causality tests with this weekly
data indicate that media Granger-causes policy, and that policy does not
Granger-cause media (using  lags; F=.; significance=.; df=,). As
with the public, then, and demonstrated in a more compelling fashion, the
media plays the dominant role in the media-policy relationship, in line with
our expectations for environmental issues.
Turning finally to results for debt/deficit, Granger tests serve to emphasize

     
the importance of the media agenda (Table C). In spite of very little autocorrelation in the media series, tests show the media has a positive effect on
both the public and policy agendas. This effect is further indicated in the
impulse response functions, which indicate that an increase in salience for the
media leads to a marked and sustained increase in the public agenda, and—while
the effects are more irregular—the policy agenda is also positively affected
(Figure G).
The expected policy effects on the media agenda are less clear—certainly,
results show this relationship is likely to be bi- rather than uni-directional.
Preliminary tests showed the two series are strongly correlated at lag ; Granger
tests offer evidence of causality running from media to Question Period and
not in the other direction. This relationship is also reflected in the impulse
response functions—there is an irregular but evident effect of media on policy,
and virtually no effect of policy on media (Figure G, I).
A more conspicuous effect of policy on the media agenda takes the form of
a strong and significant Throne Speech coefficient at lag ,13 and, consequently,
a significant Granger test for Throne Speech coefficients in Table C. The
significant effects here are likely attributable to the  Throne Speech.
Interestingly, this Throne Speech variable is not a significant predictor for the
Question Period series. Nevertheless, the speech seems to have sparked media
attention to the issue, and we can hypothesize that subsequent attention is in
large part a product of this increased media attention.
The significant Throne Speech coefficient is evidence of two facts. First, it
is clear that the debt/deficit issue was driven by both media and policymakers
in Canada. Media plays an important role in increasing attention to debt/deficit
issues, but this attention was sparked by the  Throne Speech. Secondly,
and more generally speaking, these results emphasize the potential problems
with using only one measure as an indication of the policy agenda. Granger
tests, after all, show a uni-directional effect of the media agenda on Question
Period content. Without the additional policy measures, we would miss an
important part of the picture of the dynamics surrounding the debt/deficit
issue in Canada.
ANALYSIS AND CONCLUSIONS
The preceding work is very much an argument for a particular way of looking
at politics. It is rooted in the belief that issues are an extraordinarily valuable
unit of analysis. Moreover, it suggests that an agenda-setting framework offers
the tools necessary to combine mass media studies, public opinion analysis, and
public policy research. These fields are intimately connected, and examining
13
The actual coefficients are not reported here, but are available upon request from the author.
   -  

them concurrently provides a more complete, and probably more accurate
picture of the political interactions in a society.
In spite of its methodological perspective, however, this work does present
specific findings, and these support specific conclusions. To summarize, we
review our two major hypotheses:
 Different issues display different agenda-setting dynamics: Our evidence suggests
that agenda-setting dynamics will vary in both magnitude and direction. Variance
in the direction of effects is most visibly demonstrated by the Granger tests.
The media, public, and policy agendas sometimes lead, and sometimes follow.
The magnitude of effects, on the other hand, is evident in the impulse response
functions. Graphs in the first and third columns of Figure  illustrate the varied
magnitude and longevity of media and policy effects, for instance, while the
second column shows a public agenda that does far more than simply follow
media content.
 This variance in dynamics is linked to issue attributes: Results confirm what
our descriptions of each issue suggest—differences in agenda-setting dynamics
are systematically linked to issue attributes. The first column of impulse response
functions, for instance, shows that media impact on the public is smaller for
inflation than for the environment and debt/deficit; Granger tests corroborate
the fact that the media leads for the latter two issues, but not the former. In
line with Zucker’s () obtrusiveness hypothesis, then, the potential for media
influence appears restricted for issues that the public experiences directly.
Inflation is one such issue, and the evidence above demonstrates that the
Canadian public (and policymakers) react to real-world economic conditions
rather than media content.
The media’s role is stronger for environmental and debt/deficit issues. For the
environment, effects between media, public, and policy agendas appear to be
multi-directional, suggesting that the increased salience in the late s was
not simply a product of media emphasis. Nevertheless, further analysis suggest
that the media played an especially important role. The debt/deficit issue, on
the other hand, displays very strong media effects on the public, while additional
exogenous variables suggest that the Throne Speech sparked media interest. In
this case, policymakers initiated the rise in issue salience—an important finding,
since very little work on public agenda-setting has sought to empirically connect
the policy, media, and public agendas.
In sum, the issues surveyed here suggest three different agenda-setting
dynamics—in Canada, from  to , inflation was real-world-driven,
environmental issues were media-driven, and debt/deficit issues were policydriven. This outcome confirms our hypotheses, and lends support to the more
general notion that there are important and traceable interactions between

     
media, public opinion, and policymakers. Narrowly conceived, our results
demonstrate the value of an agenda-setting framework, a means of modeling
media effects and the policy process, and the importance of issue attributes in
agenda-setting processes. Broadly conceived, they provide an empirical snapshot
of relationships between major actors in Canadian politics.
REFERENCES
Ader, C. R. (): ‘A longitudinal study of agenda setting for the issue of environmental
pollution’, Journalism and Mass Communication Quarterly, , –.
Atwater, T., Salwen, M. B., and Anderson, R. B. (): ‘Media agenda-setting with
environmental issues’, Journalism Quarterly, , –.
Bartels, L. M. (): ‘Politicians and the press: Who leads, who follows?’. Paper
Presented at the Annual Meeting of the American Political Science Association,
September , San Francisco.
Baumgartner, F. R. and Jones, B. D. (): Agendas and Instability in American Politics,
Chicago, University of Chicago Press.
Behr, R. L. and Iyengar, S. (): ‘Television news, real-world cues, and changes in
the public agenda’, Public Opinion Quarterly, , –.
Bernanke, B. (): ‘Alternative explanations of the money-income correlation’, Carnegie-Rochester Conference Series on Public Policy, , –.
Brosius, H.-B. and Kepplinger, H. M. (): ‘The agenda-setting function of television
news: Static and dynamic views’, Communication Research, , –.
Cobb, R. W. and Elder, C. D. (): Participation in American Politics: The Dynamics
of Agenda Building, Boston, Allyn and Bacon, Inc.
Cohen, B. C. (): The Press and Foreign Policy, Princeton, Princeton University
Press.
Dearing, J. W. (): ‘Setting the polling agenda for the issue of ADS’, Public Opinion
Quarterly, , –.
Dearing, J. W. and Rogers, E. M. (): Agenda Setting, Thousand Oaks, Sage.
Demers, D. P., Craff, D., Choi, Y.-H., and Pession, B. M. (): ‘Issue obtrusiveness
and the agenda-setting effects of national network news’, Communication Research,
, –.
Dodge, D. A. (): ‘Reflections on the role of fiscal policy: The Doug Purvis Memorial
Lecture’, Canadian Public Policy, , –.
Downs, A. (): ‘Up and down with ecology: The “issue attention cycle”’, The Public
Interest, , –.
Dwivedi, T. D. and Srivastava,V. K. (): ‘Optimality of least squares in the seemingly
unrelated regression equation model’, Journal of Econometrics, , –.
Enders, W. (): RATS Handbook for Econometric Time Series, New York, John Wiley
& Sons.
Freeman, J. R. (): ‘Granger causality and the times series analysis of political
relationships’, American Journal of Political Science, , –.
Freeman, J. R., Williams, J. T., and Lin, T. (): ‘Vector autoregression and the
study of politics’, American Journal of Political Science, , –.
   -  

Gonzenbach, W. J. (): The Media, the President, and Public Opinion: A Longitudinal
Analysis of the Drug Issue, –, Mahwah, NJ, Lawrence Erlbaum Associates.
Granger, C. W. J. (): ‘Investigating causal relations by econometric models and
cross-spectral methods’, Econometrica, , –.
Greene, W. H. (): Econometric Analysis, nd edn, New York, Macmillan.
Hansen, A. (): ‘The media and the social construction of the environment’, Media
Culture and Society, , –.
Hester, J. B. and Gonzenbach, W. J. (): ‘The environment: TV news, real-world
cues, and public opinion over time’, Mass Communication Review, , –.
Howlett, M. (): ‘Issue-attention and punctuated equilibria models reconsidered:
An empirical examination of the dynamics of agenda-setting in Canada’, Canadian
Journal of Political Science, , –.
Howlett, M. (): ‘Predictable and unpredictable policy windows: Institutional and
exogenous correlates of Canadian federal agenda-setting’, Canadian Journal of Political
Science, , –.
Iyengar, S. and Kinder, D. R. (): News That Matters, Chicago, University of
Chicago Press.
Jasperson, A. E., Shah, D. V., Watts, M., Faber, R. J., and Fan, D. P. (): ‘Framing
and the public agenda: Media effects on the importance of the federal budget deficit’,
Political Communication, , –.
Johnson, D. R. (): ‘Expected inflation in Canada –: An evaluation of Bank
of Canada credibility and the effect of inflation targets’, Canadian Public Policy, ,
–.
Kingdon, J. W. (): Agendas, Alternatives, and Public Policies, nd edn, New York,
Harper Collins.
Kneebone, R. D. (): ‘Deficits and debt in Canada: Some lessons from recent
history’, Canadian Public Policy, , –.
MacKuen, M. B. and Coombs, S. L. (): More Than News: Media Power in Public
Affairs, Beverly Hills, CA, Sage Publications.
Mayer, R. N. (): ‘Gone yesterday, here today: Consumer issues in the agenda-setting
process’, Journal of Social Issues, , –.
McCombs, M. E. and Shaw, D. L. (): ‘The agenda-setting function of the mass
media’, Public Opinion Quarterly, , –.
McCombs, M. E. and Shaw, D. L. (): ‘The evolution of agenda-setting research:
twenty-five years in the marketplace of ideas’, Journal of Communication, , –
.
Page, B.I. and Shapiro, R. Y. (): ‘Effects of public opinion on policy’, American
Political Science Review, , –.
Page, B. I. and Shapiro, R. Y. (): The Rational Public: Fifty Years of Trends in
Americans’ Policy Preferences, Chicago, University of Chicago Press.
Parlour, J. W. and Schatzow, S. (): ‘The mass media and public concern for
environmental problems in Canada, –’, International Journal of Environmental
Studies, , –.
Pritchard, D. (): ‘Homicide and bargained justice: the agenda-setting effect of
crime news on prosecutors’, Public Opinion Quarterly, , –.

     
Pritchard, D. and Berkowitz, D. (): ‘The limits of agenda-setting: The press and
political responses to crime in the United States, –’, International Journal
of Public Opinion Research, , –.
Rogers, E. M. and Dearing, J. W. (): ‘Agenda-setting research: where has it been,
where is it going?’. In J. A. Anderson (ed.): Communication Yearbook, Vol. , London,
Sage.
Rogers, E. M., Dearing, J. W., and Bregman, D. (): ‘The anatomy of agenda-setting
research’, Journal of Communication, , –.
Sims, C. A. (): ‘Are forecasting methods usable for policy analysis?’, Federal Reserve
Bank of Minneapolis Quarterly Review, , –.
Smith, K. A. (): ‘Newspaper coverage and public concern about community issues:
A time series analysis’, Journalism Monographs, Columbia, SC, Association for
Education in Journalism and Mass Communication.
Smith, T. W. (): ‘America’s most important problem—A trend analysis, –’,
Public Opinion Quarterly, , –.
Smith, T. W. (): ‘The polls: America’s most important problems, Part I: National
and international’, Public Opinion Quarterly, , –.
Soroka, S. N. (a): ‘Different issues, different effects: Building an issue typology
for agenda-setting’. Paper presented at the Annual General Meeting of the American
Political Science Association.
Soroka, S. N. (b): ‘Policy agenda-setting theory revisited: A critique of Howlett
on Downs, Baumgartner and Jones, and Kingdon’, Canadian Journal of Political
Science, , –.
Soroka, S. N. (): Agenda-Setting Dynamics in Canada, Vancouver, BC, University
of British Columbia Press.
Thiessen, G. G. (): ‘The Canadian experience with targets for inflation control’,
Canadian Public Policy, , –.
Wanta, W. and Hu, Y. W. (): ‘The agenda-setting effects of international news
coverage: An examination of differing news frames’, International Journal of Public
Opinion Research, , –.
Wanta, W., Stephenson, M., Turk, J. V., and McCombs, M. E. (): ‘How the
President’s State of the Union Talk influenced news media agendas’, Journalism
Quarterly, , –.
Weaver, D. H. (): ‘Issue salience and public opinion: Are there consequences of
agenda-setting?’, International Journal of Public Opinion Research, , –.
Winter, J. P., Eyal, C. H., and Rogers, A. H. (): ‘Issue specific agenda-setting:
The whole is less than the sum of the parts’, Canadian Journal of Communication, ,
–.
Yagade, A. and Dozier, D. M. (): ‘The media agenda-setting effect of concrete
versus abstract issues’, Journalism Quarterly, , –.
Zellner, A. (): ‘An efficient method of estimating seemingly unrelated regressions
and tests for aggregation bias’, Journal of the American Statistical Association, ,
–.
Zucker, H. G. (): ‘The variable nature of news media influence’. In B. D. Ruben
(ed.): Communication Yearbook, Vol. , New Brunswick, NJ, Transaction Books.
   -  

BIOGRAPHICAL NOTE
Stuart N. Soroka is currently the Gwilym Gibbon Postdoctoral Research Fellow at
Nuffield College, Oxford; as of Fall , he will be an assistant professor in the
Department of Political Science, McGill University, Montreal, Canada. His research
interests include comparative political communications and public policy.
Address correspondence to Stuart N. Soroka, Gwilym Gibbon Postdoctoral Research
Fellow, Nuffield College, New Road, Oxford, OX NF, UK, Email: stuart.soroka@
nuffield.oxford.ac.uk